diff --git a/.claude/skills/check-rs/SKILL.md b/.claude/skills/check-rs/SKILL.md
new file mode 100755
index 0000000..5f01826
--- /dev/null
+++ b/.claude/skills/check-rs/SKILL.md
@@ -0,0 +1,94 @@
+---
+name: check-rs
+description: Rust Code Quality Pass — fmt, clippy, DRY/complexity/test review agents, then fixes all findings
+user-invocable: true
+---
+
+# /check-rs — Rust Code Quality Pass
+
+Perform a comprehensive quality pass on recently changed Rust code. Fix every issue found — do not just report them.
+
+## Phase 1: Identify what changed
+
+Run `git diff HEAD --stat` and `git status --short` to identify the changed and untracked files. Exclude `Cargo.lock`. If there are no git changes, ask the user which files to review.
+
+## Phase 2: Run the tooling (fix before reviewing)
+
+Run these commands and fix every failure before proceeding. Do not skip to the agents if either tool reports problems.
+
+**Formatting:**
+```
+cargo fmt
+```
+Apply formatting automatically — do not just check.
+
+**Linting:**
+```
+cargo clippy -- -D warnings
+```
+Warnings are treated as errors. Fix every warning in the changed files (and any file you touch while fixing). Do not use `#[allow(...)]` to suppress a warning unless the suppression has a doc comment explaining why it is necessary.
+
+After fixing all clippy issues, run `cargo build` to confirm a clean compile with zero warnings.
+
+## Phase 3: Launch three review agents in parallel
+
+Use the Agent tool to launch all three agents **in a single message** (parallel). Pass each agent the list of changed files and their full content.
+
+### Agent 1: DRY + Organisation Review
+For each changed file:
+1. Search for existing utilities, helpers, and patterns elsewhere in the codebase that could replace newly written code. Check adjacent modules, utility files, and the nearest `mod.rs`.
+2. Flag any function that duplicates existing functionality and name the existing alternative.
+3. Flag inline logic that re-implements something already in the standard library or a crate already in `Cargo.toml`.
+4. Flag modules or types that belong in a different file/module based on the existing project structure.
+5. Flag any string literal that should be a named constant based on how it is used.
+
+Report: file:line, what duplicates what, the exact fix.
+
+### Agent 2: Complexity + Documentation Review
+For each changed file:
+1. **Long functions** — flag any function over ~40 lines that could be split. Suggest what the extracted helper should be called and what it should contain.
+2. **Deep nesting** — flag `match`/`if`/`for` blocks nested more than 3 levels deep. Suggest early-return or helper extraction.
+3. **Missing docs** — flag every `pub` item (fn, struct, enum, const, mod) that lacks a doc comment. Suggest the doc text based on the item's name and usage.
+4. **Useless comments** — flag comments that describe *what* the code does (well-named code already does that), narrate the change, or reference the task. Keep only *why* comments (hidden constraints, non-obvious invariants, workarounds).
+5. **Overly broad error messages** — error strings that don't identify the context (file, field, operation).
+
+Report: file:line, the issue, the exact fix.
+
+### Agent 3: Test Coverage Review
+For each changed file:
+1. List every `pub fn`, `pub async fn`, and non-trivial private fn in the file.
+2. For each function, determine whether a test exists that exercises its main path and at least one edge case (empty input, `None`, boundary value, error path).
+3. Flag functions with no test and functions with only a happy-path test.
+4. Check that the test module uses `#[cfg(test)]` and that test names describe the scenario (`test_<function>_<scenario>`).
+5. Check that pure/deterministic functions (parsers, validators, formatters) have property-style tests (multiple inputs in one test).
+
+Report: file:function, what test is missing, a concrete test to add (include the code).
+
+## Phase 4: Fix all findings
+
+Wait for all three agents to complete. Work through every finding:
+- Apply the fix directly using Edit/Write tools.
+- If a finding is a clear false positive, note it and skip.
+- Do not add `#[allow(...)]` without a doc comment explaining the suppression.
+- Do not add tests that test nothing (trivial getters, `Default::default()` calls).
+
+## Phase 5: Final verification
+
+Run in sequence:
+```
+cargo fmt --check
+cargo clippy -- -D warnings
+cargo test
+```
+
+All three must pass cleanly. If any fail, fix and re-run before declaring done.
+
+## Summary
+
+Report what was fixed in each category:
+- **Tooling** — fmt/clippy changes made
+- **DRY / Organisation** — duplications removed, constants added, code moved
+- **Complexity / Docs** — functions split, docs added, comments removed
+- **Tests** — new tests added (list them)
+
+If a category had no issues, say so explicitly.